Skip to main content
Text ready — video coming soon

SDE and SDD: connect the workflow to your judgment

Hello again! It's time to name what you've been doing. Over the last two lessons, we wrote down what the committee report must do, ran a program an AI wrote from it, and compared its results with answers we worked out first. The overpayment showed a question the spec still had to answer.

That way of working has names. In this lesson, we'll:

  • retrace the committee work as one loop of four steps;
  • meet spec-driven development, through one toolkit;
  • meet this book's wider method, and see how it differs;
  • see what our small example does and doesn't show.

Why it matters: names let you talk about a way of working, with people and with AI. Once you can name each step, you can notice when one has been skipped.

How the committee work becomes one loop​

Here is the committee work again, as four steps:

StepWhat happenedWhere you met it
SpecifyThe purpose, inputs, output, boundaries and acceptance criteria were written down firstSpecifications and design
BuildAn AI wrote the program from the spec. An agent doing this should have only the access its task needsAI coding agents
VerifyThe program's results were compared with the expected ones, and its lines were readVerify the committee example
OwnYou can explain the report line by line, and you take the overpayment question back to the organiserThis lesson

Own is the step people forget. It means you can explain what the work does and why, and you're ready to answer for it. If the organiser asks, "Why does it say 500 still due?", you can show them the calculation, line by line.

What spec-driven development (SDD) is​

Spec-driven development (SDD) is a name for ways of working in which a written spec guides the code. The spec stays the reference for both the person and the AI as the work changes. Writing requirements down and checking code against them is much older than AI; SDD applies that habit to working with AI.

GitHub released a toolkit called Spec Kit in September 2025 to support this way of working. Its description splits the work before coding into three steps:

  1. Specify: describe what you want, and why.
  2. Plan: decide the technical approach.
  3. Tasks: break the plan into small pieces of work.

After those steps, the code is written from the tasks, usually by AI.

How does that fit our loop? Spec Kit's three steps are a more detailed version of our Specify step; our design decision, for example, belongs in its Plan. Its coding step is our Build. Our Verify and Own are still needed, whatever a toolkit calls its steps. You don't need Spec Kit to start: a plain text spec like ours is enough.

What Spec-Driven Engineering (SDE) adds​

Spec-Driven Engineering, or SDE, is the method at the centre of this book. The engineer specifies the intent and the boundaries, the AI builds within them, and a human verifies the work before taking ownership of it. SDE is this book's own name for that wider method, built on SDD. When you meet similar words elsewhere, check what their author means by them.

So what's the difference?

  • SDD puts the spec at the centre of development: producing the code.
  • SDE puts engineering at the centre: the judgment around the code. That means the intent, limits that are actually enforced, checks at set points, decisions about risk, and owning the result.

Remember the two words from the Code Literacy lesson. Development is producing the code, and AI has made that much cheaper. Engineering is the judgment around it, and that part stays yours. SDE is built around it. Good SDD teams do this judgment work too; SDE makes it the centre and gives it a name.

Why a spec doesn't guarantee the code​

Our example was small: two programs written for teaching, and a few checks. It shows how a clear spec and answers worked out in advance can catch a mistake.

And a spec alone doesn't guarantee the code. A program can follow a spec's words and still surprise you, as the overpayment did. That's why verifying is part of the method, never an extra.

Check what you can now explain​

Without looking back:

  1. Name the four steps of the loop, and give the committee example for each.
  2. In one sentence each, say what SDD is for, and what SDE adds.
  3. Why does "own" mean being able to explain the work, not just run it?

When you can answer these, you can name the loop and explain it through the example. In the final lesson of this stage, we'll see how the rest of the book builds on it, and where you go next.